Algorithms for dynamical low-rank approximation

Bart Vandereycken
Université de Genève
Mathematics

The dynamical low-rank approximation or the time-dependent variational principle allows to directly approximate large matrix and tensor differential equations by low-rank matrices and tensors. A popular algorithm that implements this idea is the projector-splitting integrator of Lubich and Oseledets. In this talk, I will discuss some recent extensions of this method that include parallelism and rank adaptivity.

Presentation (PDF File)

Back to Workshop III: Mathematical Foundations and Algorithms for Tensor Computations